Title:
BEHAVIOR FEATURE EXTRACTION METHOD, SYSTEM AND APPARATUS BASED ON TIME-SPACE/FREQUENCY DOMAIN HYBRID LEARNING
Document Type and Number:
WIPO Patent Application WO/2020/113886
Kind Code:
A1
Abstract:
Disclosed are a behavior feature extraction method, system and apparatus based on time-space/frequency domain hybrid learning. The method comprises: acquiring a video behavior sequence based on a skeleton, and extracting a time-space domain behavior feature map by means of a conversion network; inputting same into a frequency domain attention network to perform frequency selection, then inverting same back to a time-space domain, and adding same to the time-space domain behavior feature map; synchronously performing local reasoning and non-local reasoning, and performing high-layer local reasoning; and globally pooling the time-space domain behavior feature map obtained by reasoning so as to obtain a behavior feature vector of the video behavior sequence.
Inventors:
HU GUYUE (CN)
CUI BO (CN)
YU SHAN (CN)
CUI BO (CN)
YU SHAN (CN)
Application Number:
PCT/CN2019/083357
Publication Date:
June 11, 2020
Filing Date:
April 19, 2019
Export Citation:
Assignee:
INST AUTOMATION CAS (CN)
International Classes:
G06K9/00; G06K9/46; G06N3/04; G06N3/08
Foreign References:
| CN107292247A | 2017-10-24 | |||
| CN108021889A | 2018-05-11 | |||
| CN108921087A | 2018-11-30 |
Attorney, Agent or Firm:
HENYOL INTELLECTUAL PROPERTY LAW CORPORATION (CN)
Download PDF:
Previous Patent: FILM LAYER STRUCTURE FOR DISPLAY APPARATUS
Next Patent: FEED PRE-COMPRESSION DEVICE FOR USE IN CROPPING MACHINE
Next Patent: FEED PRE-COMPRESSION DEVICE FOR USE IN CROPPING MACHINE